A METHOD FOR BONDING AND INTERCONNECTING SEMICONDUCTOR CHIPS

Abstract

 The invention is related to a method for bonding and interconnecting two semiconductor chips arranged on semiconductor substrates (2a,2b). HSQ (Hydrogen Silsesquioxane) or an equivalent material is used as a bonding layer and after bonding and thinning one of the wafers (or first thinning and then bonding), the bond layer (7) is locally irradiated by an e-beam through the thinned substrate (2a), thereby locally transforming the bonding material into silicon oxide. Then a via opening (12) is etched through the thinned substrate and an etch process selectively removes the oxide from an area delimited by the bonding material or vice versa. The filling of the via opening (12) establishes an electrical connection between the bonded wafers, that is equivalent to a connection obtained by hybrid bonding, but that does not suffer from the disadvantages thereof.

Description

Field of the invention

[0001] The invention is related to the 3D-integration of semiconductor integrated circuit chips, i.e. to the production of stacked interconnected chips by bonding of device wafers.

State of the art.

[0002] Various approaches for stacked 3D integration have been studied and developed. A favoured approach involves the bonding of fully processed device wafers followed by the formation of the electrical interconnection of the devices on the respective wafers by Through Semiconductor Via (TSV) connections formed by etching through a thinned device wafer. This particular process flow is often referred to as the 'TSV via-last flow'. In the TSV via-last flow, the two current bonding options are :

1) dielectric bonding, in which case uniform dielectric bonding layers are applied to the two wafers which are brought into contact to form the wafer stack, followed by an annealing step to establish the bond between the dielectric bonding layers. In this case, the two wafers are contacted from the top of the stack through different types of TSV (deep and shallow).

2) hybrid bonding, in which case the bonding interface comprises metal pads embedded in dielectric bonding layers applied on the two wafers, which enables the formation of electrical connections between the two wafers, while a single type of TSV is sufficient to connect the wafers to the top of the stack.

[0003] Option 1 (dielectric bonding) is challenging for TSV processing by lithography and etching, since it requires the etch to be compatible with both types of TSV (deep and shallow) or if this is not possible, the approach requires different reticles, which is costly. Option 2 (hybrid bonding) is challenging from the bonding perspective since two different materials are present at the interface : metal and dielectric. This may cause errors in terms of the alignment of the metal pads due to non-uniform wafer deformations occurring during the bonding process. Another problem is related to the preparation of the bonding surfaces for hybrid bonding. This is typically done by damascene processing, which includes chemical mechanical polishing (CMP), in order to obtain for example Cu conductors embedded in SiO₂. It is however difficult to obtain a very flat surface due to the Cu recess below the SiO₂ surface during CMP. As a result, the hybrid bonding surface is not perfectly flat, leading to an inferior quality of the bond.

Summary of the invention

[0004] The present invention aims to provide a method that suffers from neither of the above-described disadvantages. This aim is achieved by the methods disclosed in the appended claims. The invention is related to a method for bonding and interconnecting two semiconductor chips arranged on substrates such as silicon wafers. The invention takes advantage of a number of known characteristics of the class of compounds known as Hydrogen Silsesquioxane (HSQ) and materials equivalent to HSQ, in order to solve the above shortcomings of the prior art. HSQ or its equivalent is used as a bonding layer for bonding two device wafers, and after bonding and thinning one of the wafers (or first thinning and then bonding), the bond layer is locally irradiated by an e-beam through the thinned wafer, thereby locally transforming the bonding material into silicon oxide. Then a via opening is etched through the thinned wafer and an etch process selectively removes the oxide from an area delimited by the bonding material or vice versa. The filling of the via opening establishes an electrical connection between the bonded wafers, that is equivalent to a connection obtained by hybrid bonding, but that does not suffer from the disadvantages thereof.

[0005] According to a first embodiment, the invention is in particular related to a method for bonding a first semiconductor chip on a first substrate to a second semiconductor chip on a second substrate, comprising the steps of :

• providing a bonding layer on at least one of the substrates, the bonding layer being formed of a dielectric bonding material, wherein the dielectric material is Hydrogen Silsesquioxane (HSQ), or an equivalent thereof,
• bonding the first substrate to the second substrate and performing a thermal annealing step, so as to obtain a stack of the substrates with a bond layer between the substrates,

characterized in that:

• the first substrate is thinned either before or after the bonding step,
• the thermal annealing step takes place at a temperature below a temperature at which the bonding material transforms into silicon oxide,
• after bonding, an electron beam is scanned across an area of the thinned substrate, the energy of the electron beam being sufficient for the beam to reach the bond layer and to transform the scanned bonding material of the bond layer into a silicon oxide, to thereby obtain a volume of the bond layer that is transformed into a silicon oxide, wherein said volume is essentially a closed wall extending over the full thickness of the bond layer, and enclosing a volume of bonding material, thereby separating that volume of bonding material from the rest of the bond layer,
• a via opening is etched through the thinned substrate until reaching the interior of the closed wall, and a dielectric liner is formed on the sidewalls of the via opening,
• through the via opening, the bonding material is removed from the interior of the closed wall, selectively with respect to the silicon oxide, so as to create a cavity in the bond layer and expose a conductor in the second substrate,
• the cavity and the via opening are filled with an electrically conductive material, to thereby form an electrical connection that connects the conductor in the second substrate to a conductor present in the first substrate and/or to a contact present on top of the stack.

[0006] After the removal of the bonding material from the interior of the closed wall, the stack may be subjected to an additional thermal annealing step at a temperature at or above the temperature at which the bonding material of the bond layer transforms into silicon oxide, to thereby transform the entire bond layer into silicon oxide.

[0007] According to a second embodiment, the invention is related to a method for bonding a first semiconductor chip on a first substrate to a second semiconductor chip on a second substrate, comprising the steps of :

• providing a bonding layer on at least one of the substrates, the bonding layer being formed of a dielectric material, wherein the dielectric material is Hydrogen Silsesquioxane (HSQ), or an equivalent thereof,
• bonding the first substrate to the second substrate, and performing a thermal annealing step, so as to obtain a stack of the substrates with a bond layer between the substrates,

characterized in that:

• the first substrate is thinned either before or after the bonding step,
• the thermal annealing step takes place at a temperature below a temperature at which the bonding material transforms into silicon oxide,
• after bonding, an electron beam is scanned across an area of the thinned substrate, the energy of the electron beam being sufficient for the beam to reach the bond layer and to transform the bonding material of the bond layer into a silicon oxide, to thereby obtain a volume of the bond layer that is transformed into a silicon oxide, wherein said volume is a solid shape extending over the full thickness of the bond layer,
• a via opening is etched through the thinned substrate until reaching the solid shape, and a dielectric liner is formed on the sidewalls of the via opening,
• through the via opening, the silicon oxide of the solid shape is removed selectively with respect to the bonding material, so as to create a cavity in the bond layer and expose a conductor in the second substrate,
• the cavity and the via opening are filled with an electrically conductive material, to thereby form an electrical connection that connects the conductor in the second substrate to a conductor in the first substrate and/or to a contact on top of the stack.

[0008] After the removal of the silicon oxide of the solid shape, the stack may be subjected to an additional thermal annealing step at a temperature at or above the temperature at which the bonding material of the bond layer transforms into silicon oxide, to thereby transform the entire bond layer into silicon oxide.

[0009] The method according to the second embodiment may further comprise the step of applying a continuous layer onto the first substrate prior to applying the bonding layer to said first substrate, wherein an additional liner is applied on top of the liner inside the via opening, and wherein the material of said continuous layer and of the additional liner is resistant to an etch process used to remove the silicon oxide of the solid shape. The continuous layer and the additional liner may be formed of silicon nitride.

[0010] According to the first or the second embodiment, a bonding layer may be provided on the two substrates, wherein the bonding comprises a dielectric bonding step obtained by bringing the two bonding layers into physical contact, followed by the first thermal annealing step.

[0011] The invention is also related to a stack of two or more semiconductor chips bonded by a bond layer consisting of Hydrogen Silsesquioxane (HSQ), or an equivalent thereof, wherein at least one electrically conductive pillar extends between two adjacent chips of the stack, and wherein said pillar is embedded in the bond layer formed between said adjacent chips.

[0012] The electrically conductive pillar may form the interior of a closed silicon oxide wall extending between the adjacent chips, said wall separating the bond layer from the pillar. Alternatively, the pillar may be in direct contact with the bond layer.

Brief description of the figures

[0013] 

Figures 1a to 1h illustrate the steps of the method for bonding and interconnecting two semiconductor chips according to a first embodiment of the invention.

Figure 2 illustrates the result of the method of the first embodiment applied to a case where two conductors in the back end of line of the two bonded chips are to be connected directly.

Figures 3a to 3e illustrate the steps of the method for bonding and interconnecting two semiconductor chips according to a second embodiment of the invention.

Figure 4 illustrates the result of the method of the second embodiment applied to a case where two conductors in the back end of line of the two bonded chips are to be connected directly.

Detailed description of the invention

[0014] The term HSQ represents a class of inorganic compounds with the chemical formula [HSiO_3/2]_n. Any compound within this class is applicable in the invention. Materials applicable instead of HSQ and in the same way as HSQ are partially condensed silsesquioxanes in which 2 or more long chains of RSiO_3/2 units are connected at regular intervals by Si-O-Si bonds, where R can be a hydrogen atom, an alkyl, an aryl or an alkoxy functional group. These alternative materials are referred to within the present context as materials equivalent to HSQ. The use of HSQ as a resist in e-beam lithography is well known and based on the fact that HSQ converts to a silicon oxide (SiO_x) after exposure to irradiation by an electron beam. It is known also that HSQ transforms into silicon oxide under the influence of heating above the curing temperature of the material, which is at about 400°C. These characteristics of HSQ are documented for example in "Comparative study of thermally cured and electron-beam-exposed hydrogen silsesquioxane resists", Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures Processing, Measurement, and Phenomena 26, 1654 (2008). It is known also that HSQ may be used as an adhesive bonding material, as described for example in document WO2010/141351. The method of the invention will be explained on the basis of the use of HSQ, but the description is valid also for the equivalent materials.

[0015] Figure 1a shows the starting point of the method according to a first embodiment of the invention for bonding and interconnecting two semiconductor chips. Two wafers 1a and 1b are to be bonded. Each of the wafers comprises a semiconductor support substrate 2a and 2b, e.g. a silicon substrate, carrying a plurality of chips, which are composed of multiple active devices (transistors, diodes and/or others) arranged on the substrate according to the layout of the chips. The drawing shows a schematic cross section of a small area of the wafers 1a and 1b as they face each other before bonding. The active devices are arranged in a front end of line (FEOL) portion on the respective substrates 2a and 2b, with on top of the FEOL portion a back end of line (BEOL) portion consisting of multiple metallization levels M₁,M₂, etc interconnected by via connections, for connecting the active devices to contact terminals on the chip or on the eventual chip stack. The combined FEOL and BEOL portion is represented in a simplified way by the areas 3a and 3b for the respective wafers 1a and 1b. Electrical conductors 4a and 4b are shown at the outer surface of the respective areas 3a and 3b, embedded in respective dielectric layers 5a and 5b. These layers 5a and 5b represent the upper level of the BEOL which may be the actual upper metallization layer M_n or a redistribution layer formed thereon. For the sake of explaining the invention, the conductors 4a/4b and outer dielectric layers 5a/5b are shown at a thickness that may be out of proportion to the real-life thickness of the FEOL and BEOL portions in a chip.

[0016] According to the invention, HSQ bonding layers 6a and 6b are applied on both wafers. These layers may be applied by spin coating an HSQ-containing solution onto the surface, followed by a baking step to remove the solvent, so that an essentially pure HSQ layer remains. HSQ-solutions are commercially available products and details of the above-described application process are known as such in the art. The thickness of the dried HSQ layers is preferably in the range of 100 to 500 nm. The wafers are aligned and the bonding layers 6a and 6b are brought into physical contact, followed by a thermal annealing step, at a temperature below the temperature at which HSQ transforms into silicon oxide. For example the post-bond anneal may be done at a temperature between 250° and 350°C during a time between 10 minutes and 2 hours At these temperatures, the two HSQ layers merge to form a single HSQ bond layer 7.

[0017] Then, as shown in Figure 1c, the top wafer 1a is thinned, which may be done by grinding and/or chemical mechanical polishing (CMP) and possibly by etching, until the support substrate 2a is reduced to a thickness which is preferably in the order of a few micrometers, for example about 2 or 3 micrometers. In the next step (Fig. 1d), the wafer assembly is irradiated with an electron beam 10 directed at the thinned substrate 2a and oriented essentially perpendicularly with respect to plane of the wafer assembly. The beam energy is selected so that the electrons penetrate the thinned upper substrate 2a and the BEOL/FEOL portion 3a, and interact with the HSQ bond layer 7 across the full thickness of the bond layer 7. The energy is furthermore chosen so as to be sufficient for transforming the HSQ material locally into silicon oxide. When the silicon substrate 2a is thinned to a few micrometers, this may be achieved by an e-beam of which the energy is in the order of 100 keV. According to the first embodiment and as shown in Figure 1d, the e-beam is scanned over a ringshaped area, so that the transformation of HSQ to silicon oxide is limited to a volume that essentially corresponds to a closed wall 11. The cross-section of the wall need not be circular as shown in the drawing, but may have other shapes as well. The height of the wall equals the thickness of the bond layer 7, so that the interior of the wall forms a volume 9 of HSQ material that is separated from the remainder of the bond layer 7 by the closed silicon oxide wall 11. The location of the irradiated area is chosen with respect to a conductor 4b in the lower substrate, so that the enclosed volume 9 overlies the conductor 4b.

[0018] By known lithography and anisotropic etching techniques and as illustrated in Figure 1e, a via opening 12 is then produced through the thinned upper substrate 2a and through the BEOL/FEOL portion 3a. In the drawing, the via opening 12 is concentric with the closed wall 11, but this is not a requirement. In general, the section of the via opening 12 as seen in a plane parallel to the wafer assembly, lies within the outer border of the wall 11 and overlaps at least partially the section of the HSQ volume 9 inside the wall 11 of silicon oxide. The etching of the via opening 12 continues until reaching said HSQ volume 9. Then a dielectric liner 13 is produced on the sidewalls and the bottom of the via opening, which may be done by a known process, for example by atomic layer deposition. The liner 13 may be silicon oxide. The liner is removed from the bottom of the opening, while maintaining the liner on the sidewalls, which may be done by a plasma-based etch process, as described for example in document EP3035369. The dimensions of the via opening 12 and the liner may be in accordance with known TSV-technology.

[0019] Then, as shown in Figure 1f, the HSQ material of the enclosed volume 9 is removed by an etch process that removes the HSQ selectively with respect to the silicon oxide wall 11 and with respect to the liner 13 and the dielectric layer 5a of the BEOL portion on the thinned substrate 2a. An suitable etch process for this step is a wet etch using a TMAH solution (Tetramethylammonium Hydroxide). A cavity 14 is thereby created, that is connected to the upper surface of the wafer stack by the via opening 12. The cavity 14 and the via opening 12 are then filled with an electrically conductive material, such as copper or tungsten, see Figure 1g, by a suitable process such as electroplating for Cu or LPCVD (Low Pressure Chemical Vapour Deposition) for W. This creates an electrical connection to the conductor 4b of the lower substrate 2a, consisting of an electrically conductive pillar 15 and a via connection 16. A blanket layer 17 of the conductive material is formed on the upper surface of the wafer stack, which may be patterned as shown in Figure 1h, to form a contact pad 18 that allows to contact a chip on the lower wafer 1b from the top of the stack. The result is a stack of two semiconductor chips 8a and 8b of which a small portion is shown in Figure 1h. The chips are bonded by an HSQ bond layer 7, wherein at least one electrically conductive pillar 15 extends between the two adjacent chips 8a and 8b of the stack, and wherein said pillar 15 is embedded in the bond layer 7 formed between said adjacent chips, but separated from the bond layer 7 by a closed silicon oxide wall 11 extending between the adjacent chips.

[0020] Figure 2 illustrates how the same process allows to form a contact to both the upper and the lower chip of a stack, by irradiating a ringshaped area that overlaps a conductor 4a on the upper substrate 2a as well as a conductor 4b on the lower substrate 2b. In both cases, Figure 1h and Figure 2, the result is equivalent to hybrid bonding wherein two metal pads formed on the respective bonding surfaces are in contact at the bonding interface. The method of the invention however does not suffer from the disadvantages of hybrid bonding, as the bonding process is in fact a direct dielectric bonding process that establishes a bond between two uniform layers, in this case HSQ layers. So the method of the invention combines the advantages of the two existing options highlighted in the introduction, without incurring the respective disadvantages.

[0021] A second embodiment of the method of the invention is illustrated in Figures 3a to 3e. The same wafers 1a and 1b are being bonded by HSQ bonding layers 6a and 6b, but an additional dielectric layer 20 is formed on the upper substrate 2a before bonding, as seen in Figure 3a. This may be a SiN layer, for reasons described further in this paragraph. The HSQ bonding layer 6a is applied on top of the SiN layer 20. The steps of bonding and thinning the upper substrate 2a are the same as described for the first embodiment. However, as seen in Figure 3b, an e-beam 10 is now scanned across a full circular area (or another shape, the circular shape is not a limitation to the invention), transforming the HSQ material into silicon oxide so as to form a solid oxide shape 21 that spans across the full thickness of the HSQ bond layer 7, and that overlies a conductor 4b in the lower substrate 2b. The via opening 12 is etched above the oxide shape 21 (see Fig. 3c), and the liner 13 is formed on the sidewalls and removed from the bottom of the via opening as in the first embodiment. Again, the via opening 12 need not be concentric with the shape 21. It is sufficient in this case that the cross-section of the via opening 12 at least partially overlaps the cross-section of the shape 21. An additional SiN liner 22 is then formed on the first liner 13, for example by plasma-enhanced atomic layer deposition, possibly according to a similar process as applied for the formation of the first liner 13, i.e. deposition of SiN on the sidewalls and bottom of the via opening and removal of SiN from the bottom. Then, as seen in Figure 3d the silicon oxide shape 21 is removed by an etching process, for example a C₄F₈ based plasma etch, that removes the silicon oxide selectively with respect to the HSQ 7 and the SiN 20+22. The function of the SiN layers is precisely to protect the liner 13 and the dielectric layer 5a from being attacked by this etch process. The result is again a cavity 23 connected to the upper surface of the stack by the via opening 12. The cavity 23 and the via opening 12 are then filled with an electrically conductive material, as illustrated in Figure 3e, again forming the pillar 15, via connection 16 and contact pad 18. The result is again a stack of two semiconductor chips 8a and 8b of which a small portion is shown in Figure 3e. The chips are bonded by an HSQ bond layer 7, wherein at least one electrically conductive pillar 15 extends between the two adjacent chips 8a and 8b of the stack, and wherein said pillar 15 is embedded in and in direct contact with the bond layer 7 formed between said adjacent chips. The SiN layer 20 and the additional SiN liner 22 may be omitted if the material of the liner 13 and of the dielectric layer 5a are themselves resistant to the etch process used for removing the silicon oxide of the shape 21.

[0022] Figure 4 illustrates how the same process allows to form a contact to both the upper and the lower chip of a stack, by irradiating an area that overlaps a conductor 4a on the upper substrate 2a as well as a conductor 4b on the lower substrate 2b.

[0023] In any of the above-described embodiments, the HSQ (or equivalent) bond layer 7 may be maintained as the final bond layer between the wafers of the stack, as is the case in the process illustrated in the drawings. Alternatively the HSQ bond layer 7 may be transformed in its entirety into silicon oxide, after the formation of the cavity 14 or 23, by a thermal anneal, for example at a temperature of about 420°C. Whether this step is included or not may depend on the thermal budget of later process steps. When the HSQ or its equivalent is not transformed by a thermal anneal, the stack of semiconductor chips obtained by the method of the invention is characterized by the presence of the pillar 15 between adjacent chips, and embedded in a bond layer of HSQ or equivalent material. When the stack is obtained by the method of the first embodiment, the pillar is separated from the HSQ bond layer 7 by the silicon oxide closed wall 11 (as in Figures 1h and 2). When obtained by the method according to the second embodiment, the pillar 15 is in direct contact with the surrounding HSQ bond layer 7 (as in Figures 3e and 4).

[0024] In the above-described process flow, thinning of the first substrate 2a is done after bonding. Alternatively, the substrate 2a could be thinned before bonding, by first temporarily bonding the wafer 1a to an auxiliary support wafer and thinning the wafer 1a (and thereby the substrate 2a) while it is temporarily attached thereto. The thinned wafer is then bonded to the second wafer 1b and the above-described process is pursued.

[0025] An HSQ or equivalent bonding layer is preferably applied on both of the surfaces that are to bonded, as was the case for the above-described embodiments (bonding layers 6a and 6b). The invention is however also applicable when an HSQ or equivalent bonding layer is applied to only one of the two surfaces which are to be bonded. For example, in the embodiments described above, the lower HSQ layer 6b may be omitted, so that the lower bonding surface is formed by the dielectric layer 5b having conductors 4b embedded therein. The bond layer 7 obtained after bonding and annealing is then formed by the single HSQ bonding layer 6a, and the methods as described above can be applied as such. The bonding then no longer qualifies as 'dielectric' bonding, but rather as a form of hybrid bonding with a continuous dielectric layer on one side of the interface and a layer comprising both metal and dielectric on the other side. This embodiment may therefore still suffer from the misalignment and other problems occurring in hybrid bonding. However, as long as these issues are kept under control, the invention is perfectly applicable in this case.

[0026] While the invention has been illustrated and described in detail in the drawings and foregoing description, such illustration and description are to be considered illustrative or exemplary and not restrictive. Other variations to the disclosed embodiments can be understood and effected by those skilled in the art in practicing the claimed invention, from a study of the drawings, the disclosure and the appended claims. In the claims, the word "comprising" does not exclude other elements or steps, and the indefinite article "a" or "an" does not exclude a plurality. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage. Any reference signs in the claims should not be construed as limiting the scope.

[0027] Unless specifically specified, the description of a layer being present, deposited or produced 'on' another layer or substrate, includes the options of

• said layer being present, produced or deposited directly on, i.e. in physical contact with, said other layer or substrate, and
• said layer being present, produced or deposited on one or a stack of intermediate layers between said layer and said other layer or substrate.

Claims

1. A method for bonding a first semiconductor chip on a first substrate (2a) to a second semiconductor chip on a second substrate (2b), comprising the steps of :

- providing a bonding layer (6a,6b) on at least one of the substrates, the bonding layer being formed of a dielectric bonding material, wherein the dielectric material is Hydrogen Silsesquioxane (HSQ), or an equivalent thereof,

- bonding the first substrate to the second substrate and performing a thermal annealing step, so as to obtain a stack of the substrates with a bond layer (7) between the substrates,

characterized in that:

- the first substrate (2a) is thinned either before or after the bonding step,

- the thermal annealing step takes place at a temperature below a temperature at which the bonding material transforms into silicon oxide,

- after bonding, an electron beam (10) is scanned across an area of the thinned substrate, the energy of the electron beam being sufficient for the beam to reach the bond layer (7) and to transform the scanned bonding material of the bond layer into a silicon oxide, to thereby obtain a volume (11) of the bond layer that is transformed into a silicon oxide, wherein said volume is essentially a closed wall (11) extending over the full thickness of the bond layer, and enclosing a volume of bonding material, thereby separating that volume of bonding material from the rest of the bond layer (7),

- a via opening (12) is etched through the thinned substrate (2a) until reaching the interior of the closed wall (11), and a dielectric liner (13) is formed on the sidewalls of the via opening (12),

- through the via opening, the bonding material is removed from the interior (9) of the closed wall (11), selectively with respect to the silicon oxide, so as to create a cavity (14) in the bond layer and expose a conductor (4b) in the second substrate,

- the cavity and the via opening (12) are filled with an electrically conductive material, to thereby form an electrical connection that connects the conductor (4b) in the second substrate to a conductor (4a) present in the first substrate and/or to a contact (18) present on top of the stack.

2. The method according to claim 1, wherein after the removal of the bonding material from the interior of the closed wall (11), the stack is subjected to an additional thermal annealing step at a temperature at or above the temperature at which the bonding material of the bond layer (7) transforms into silicon oxide, to thereby transform the entire bond layer (7) into silicon oxide.

3. A method for bonding a first semiconductor chip on a first substrate (2a) to a second semiconductor chip on a second substrate (2b), comprising the steps of :

- providing a bonding layer (6a,6b) on at least one of the substrates, the bonding layer being formed of a dielectric material, wherein the dielectric material is Hydrogen Silsesquioxane (HSQ), or an equivalent thereof,

- bonding the first substrate to the second substrate, and performing a thermal annealing step, so as to obtain a stack of the substrates with a bond layer (7) between the substrates,

characterized in that:

- the first substrate (2a) is thinned either before or after the bonding step,

- the thermal annealing step takes place at a temperature below a temperature at which the bonding material transforms into silicon oxide,

- after bonding, an electron beam (10) is scanned across an area of the thinned substrate, the energy of the electron beam being sufficient for the beam to reach the bond layer (7) and to transform the bonding material of the bond layer into a silicon oxide, to thereby obtain a volume (11) of the bond layer that is transformed into a silicon oxide, wherein said volume is a solid shape (21) extending over the full thickness of the bond layer (7),

- a via opening (12) is etched through the thinned substrate (2a) until reaching the solid shape (21), and a dielectric liner (13) is formed on the sidewalls of the via opening (12),

- through the via opening, the silicon oxide of the solid shape (21) is removed selectively with respect to the bonding material, so as to create a cavity (23) in the bond layer (7) and expose a conductor (4b) in the second substrate,

- the cavity (23) and the via opening (12) are filled with an electrically conductive material, to thereby form an electrical connection that connects the conductor (4b) in the second substrate to a conductor (4a) in the first substrate and/or to a contact (18) on top of the stack.

4. The method according to claim 3, wherein after the removal of the silicon oxide of the solid shape (21), the stack is subjected to an additional thermal annealing step at a temperature at or above the temperature at which the bonding material of the bond layer (7) transforms into silicon oxide, to thereby transform the entire bond layer (7) into silicon oxide.

5. The method according to claim 3 or 4, comprising the step of applying a continuous layer (20) onto the first substrate (2a) prior to applying the bonding layer (6a) to said first substrate, and wherein an additional liner (22) is applied on top of the liner (13) inside the via opening (12), wherein the material of said continuous layer (20) and of the additional liner (22) is resistant to an etch process used to remove the silicon oxide of the solid shape (21).

6. The method according to claim 5, wherein the continuous layer (20) and the additional liner (22) are formed of silicon nitride.

7. The method according to any one of the preceding claims, wherein a bonding layer (6a,6b) is provided on the two substrates (2a,2b), and wherein the bonding comprises a dielectric bonding step obtained by bringing the two bonding layers (6a,6b) into physical contact, followed by the first thermal annealing step.

8. A stack of two or more semiconductor chips (8a,8b) bonded by a bond layer (7) consisting of Hydrogen Silsesquioxane (HSQ), or an equivalent thereof, wherein at least one electrically conductive pillar (15) extends between two adjacent chips (8a,8b) of the stack, and wherein said pillar is embedded in the bond layer (7) formed between said adjacent chips.

9. The stack of chips according to claim 8, wherein the electrically conductive pillar (15) forms the interior of a closed silicon oxide wall (11) extending between the adjacent chips (8a,8b), said wall separating the bond layer (7) from the pillar (15).

10. The stack of chips according to claim 8, wherein the pillar (15) is in direct contact with the bond layer (7).

Drawing